4 files changed, 227 insertions, 1 deletions

diff --git a/src/video_core/debug_utils/debug_utils.cpp b/src/video_core/debug_utils/debug_utils.cpp
index ac895ec3a..f41249eac 100644
--- a/src/video_core/debug_utils/debug_utils.cpp
+++ b/src/video_core/debug_utils/debug_utils.cpp
@@ -2,6 +2,7 @@
 // Licensed under GPLv2
 // Refer to the license.txt file included.
 
+#include <algorithm>
 #include <fstream>
 #include <string>
 
@@ -55,6 +56,210 @@ void GeometryDumper::Dump() {
     }
 }
 
+#pragma pack(1)
+struct DVLBHeader {
+    enum : u32 {
+        MAGIC_WORD = 0x424C5644, // "DVLB"
+    };
+
+    u32 magic_word;
+    u32 num_programs;
+//    u32 dvle_offset_table[];
+};
+static_assert(sizeof(DVLBHeader) == 0x8, "Incorrect structure size");
+
+struct DVLPHeader {
+    enum : u32 {
+        MAGIC_WORD = 0x504C5644, // "DVLP"
+    };
+
+    u32 magic_word;
+    u32 version;
+    u32 binary_offset;  // relative to DVLP start
+    u32 binary_size_words;
+    u32 swizzle_patterns_offset;
+    u32 swizzle_patterns_num_entries;
+    u32 unk2;
+};
+static_assert(sizeof(DVLPHeader) == 0x1C, "Incorrect structure size");
+
+struct DVLEHeader {
+    enum : u32 {
+        MAGIC_WORD = 0x454c5644, // "DVLE"
+    };
+
+    enum class ShaderType : u8 {
+        VERTEX = 0,
+        GEOMETRY = 1,
+    };
+
+    u32 magic_word;
+    u16 pad1;
+    ShaderType type;
+    u8 pad2;
+    u32 main_offset_words; // offset within binary blob
+    u32 endmain_offset_words;
+    u32 pad3;
+    u32 pad4;
+    u32 constant_table_offset;
+    u32 constant_table_size; // number of entries
+    u32 label_table_offset;
+    u32 label_table_size;
+    u32 output_register_table_offset;
+    u32 output_register_table_size;
+    u32 uniform_table_offset;
+    u32 uniform_table_size;
+    u32 symbol_table_offset;
+    u32 symbol_table_size;
+
+};
+static_assert(sizeof(DVLEHeader) == 0x40, "Incorrect structure size");
+#pragma pack()
+
+void DumpShader(const u32* binary_data, u32 binary_size, const u32* swizzle_data, u32 swizzle_size,
+                u32 main_offset, const Regs::VSOutputAttributes* output_attributes)
+{
+    // NOTE: Permanently enabling this just trashes hard disks for no reason.
+    //       Hence, this is currently disabled.
+    return;
+
+    struct StuffToWrite {
+        u8* pointer;
+        u32 size;
+    };
+    std::vector<StuffToWrite> writing_queue;
+    u32 write_offset = 0;
+
+    auto QueueForWriting = [&writing_queue,&write_offset](u8* pointer, u32 size) {
+        writing_queue.push_back({pointer, size});
+        u32 old_write_offset = write_offset;
+        write_offset += size;
+        return old_write_offset;
+    };
+
+    // First off, try to translate Pica state (one enum for output attribute type and component)
+    // into shbin format (separate type and component mask).
+    union OutputRegisterInfo {
+        enum Type : u64 {
+            POSITION = 0,
+            COLOR = 2,
+            TEXCOORD0 = 3,
+            TEXCOORD1 = 5,
+            TEXCOORD2 = 6,
+        };
+
+        BitField< 0, 64, u64> hex;
+
+        BitField< 0, 16, Type> type;
+        BitField<16, 16, u64> id;
+        BitField<32,  4, u64> component_mask;
+    };
+
+    // This is put into a try-catch block to make sure we notice unknown configurations.
+    std::vector<OutputRegisterInfo> output_info_table;
+        for (int i = 0; i < 7; ++i) {
+            using OutputAttributes = Pica::Regs::VSOutputAttributes;
+
+            // TODO: It's still unclear how the attribute components map to the register!
+            //       Once we know that, this code probably will not make much sense anymore.
+            std::map<OutputAttributes::Semantic, std::pair<OutputRegisterInfo::Type, u32> > map = {
+                { OutputAttributes::POSITION_X, { OutputRegisterInfo::POSITION, 1} },
+                { OutputAttributes::POSITION_Y, { OutputRegisterInfo::POSITION, 2} },
+                { OutputAttributes::POSITION_Z, { OutputRegisterInfo::POSITION, 4} },
+                { OutputAttributes::POSITION_W, { OutputRegisterInfo::POSITION, 8} },
+                { OutputAttributes::COLOR_R, { OutputRegisterInfo::COLOR, 1} },
+                { OutputAttributes::COLOR_G, { OutputRegisterInfo::COLOR, 2} },
+                { OutputAttributes::COLOR_B, { OutputRegisterInfo::COLOR, 4} },
+                { OutputAttributes::COLOR_A, { OutputRegisterInfo::COLOR, 8} },
+                { OutputAttributes::TEXCOORD0_U, { OutputRegisterInfo::TEXCOORD0, 1} },
+                { OutputAttributes::TEXCOORD0_V, { OutputRegisterInfo::TEXCOORD0, 2} },
+                { OutputAttributes::TEXCOORD1_U, { OutputRegisterInfo::TEXCOORD1, 1} },
+                { OutputAttributes::TEXCOORD1_V, { OutputRegisterInfo::TEXCOORD1, 2} },
+                { OutputAttributes::TEXCOORD2_U, { OutputRegisterInfo::TEXCOORD2, 1} },
+                { OutputAttributes::TEXCOORD2_V, { OutputRegisterInfo::TEXCOORD2, 2} }
+            };
+
+            for (const auto& semantic : std::vector<OutputAttributes::Semantic>{
+                                                output_attributes[i].map_x,
+                                                output_attributes[i].map_y,
+                                                output_attributes[i].map_z,
+                                                output_attributes[i].map_w     }) {
+                if (semantic == OutputAttributes::INVALID)
+                    continue;
+
+                try {
+                    OutputRegisterInfo::Type type = map.at(semantic).first;
+                    u32 component_mask = map.at(semantic).second;
+
+                    auto it = std::find_if(output_info_table.begin(), output_info_table.end(),
+                                        [&i, &type](const OutputRegisterInfo& info) {
+                                            return info.id == i && info.type == type;
+                                        }
+                                        );
+
+                    if (it == output_info_table.end()) {
+                        output_info_table.push_back({});
+                        output_info_table.back().type = type;
+                        output_info_table.back().component_mask = component_mask;
+                        output_info_table.back().id = i;
+                    } else {
+                        it->component_mask = it->component_mask | component_mask;
+                    }
+                } catch (const std::out_of_range& oor) {
+                    _dbg_assert_msg_(GPU, 0, "Unknown output attribute mapping");
+                    ERROR_LOG(GPU, "Unknown output attribute mapping: %03x, %03x, %03x, %03x",
+                              (int)output_attributes[i].map_x.Value(),
+                              (int)output_attributes[i].map_y.Value(),
+                              (int)output_attributes[i].map_z.Value(),
+                              (int)output_attributes[i].map_w.Value());
+                }
+            }
+        }
+
+
+    struct {
+        DVLBHeader header;
+        u32 dvle_offset;
+    } dvlb{ {DVLBHeader::MAGIC_WORD, 1 } }; // 1 DVLE
+
+    DVLPHeader dvlp{ DVLPHeader::MAGIC_WORD };
+    DVLEHeader dvle{ DVLEHeader::MAGIC_WORD };
+
+    QueueForWriting((u8*)&dvlb, sizeof(dvlb));
+    u32 dvlp_offset = QueueForWriting((u8*)&dvlp, sizeof(dvlp));
+    dvlb.dvle_offset = QueueForWriting((u8*)&dvle, sizeof(dvle));
+
+    // TODO: Reduce the amount of binary code written to relevant portions
+    dvlp.binary_offset = write_offset - dvlp_offset;
+    dvlp.binary_size_words = binary_size;
+    QueueForWriting((u8*)binary_data, binary_size * sizeof(u32));
+
+    dvlp.swizzle_patterns_offset = write_offset - dvlp_offset;
+    dvlp.swizzle_patterns_num_entries = swizzle_size;
+    u32 dummy = 0;
+    for (int i = 0; i < swizzle_size; ++i) {
+        QueueForWriting((u8*)&swizzle_data[i], sizeof(swizzle_data[i]));
+        QueueForWriting((u8*)&dummy, sizeof(dummy));
+    }
+
+    dvle.main_offset_words = main_offset;
+    dvle.output_register_table_offset = write_offset - dvlb.dvle_offset;
+    dvle.output_register_table_size = output_info_table.size();
+    QueueForWriting((u8*)output_info_table.data(), output_info_table.size() * sizeof(OutputRegisterInfo));
+
+    // TODO: Create a label table for "main"
+
+
+    // Write data to file
+    static int dump_index = 0;
+    std::string filename = std::string("shader_dump") + std::to_string(++dump_index) + std::string(".shbin");
+    std::ofstream file(filename, std::ios_base::out | std::ios_base::binary);
+
+    for (auto& chunk : writing_queue) {
+        file.write((char*)chunk.pointer, chunk.size);
+    }
+}
+
 } // namespace
 
 } // namespace
diff --git a/src/video_core/debug_utils/debug_utils.h b/src/video_core/debug_utils/debug_utils.h
index 9b4dce539..bd7a0a89b 100644
--- a/src/video_core/debug_utils/debug_utils.h
+++ b/src/video_core/debug_utils/debug_utils.h
@@ -35,6 +35,9 @@ private:
     std::vector<Face> faces;
 };
 
+void DumpShader(const u32* binary_data, u32 binary_size, const u32* swizzle_data, u32 swizzle_size,
+                u32 main_offset, const Regs::VSOutputAttributes* output_attributes);
+
 } // namespace
 
 } // namespace
diff --git a/src/video_core/pica.h b/src/video_core/pica.h
index 640830144..fe886c16f 100644
--- a/src/video_core/pica.h
+++ b/src/video_core/pica.h
@@ -57,7 +57,7 @@ struct Regs {
 
     INSERT_PADDING_WORDS(0x1);
 
-    union {
+    union VSOutputAttributes {
         // Maps components of output vertex attributes to semantics
         enum Semantic : u32
         {
diff --git a/src/video_core/vertex_shader.cpp b/src/video_core/vertex_shader.cpp
index 93830a96a..8df14b51f 100644
--- a/src/video_core/vertex_shader.cpp
+++ b/src/video_core/vertex_shader.cpp
@@ -4,6 +4,7 @@
 
 #include "pica.h"
 #include "vertex_shader.h"
+#include "debug_utils/debug_utils.h"
 #include <core/mem_map.h>
 #include <common/file_util.h>
 
@@ -50,6 +51,11 @@ struct VertexShaderState {
     };
     u32 call_stack[8]; // TODO: What is the maximal call stack depth?
     u32* call_stack_pointer;
+
+    struct {
+        u32 max_offset; // maximum program counter ever reached
+        u32 max_opdesc_id; // maximum swizzle pattern index ever used
+    } debug;
 };
 
 static void ProcessShaderCode(VertexShaderState& state) {
@@ -57,6 +63,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
         bool increment_pc = true;
         bool exit_loop = false;
         const Instruction& instr = *(const Instruction*)state.program_counter;
+        state.debug.max_offset = std::max<u32>(state.debug.max_offset, 1 + (state.program_counter - shader_memory));
 
         const float24* src1_ = (instr.common.src1 < 0x10) ? state.input_register_table[instr.common.src1]
                              : (instr.common.src1 < 0x20) ? &state.temporary_registers[instr.common.src1-0x10].x
@@ -88,6 +95,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
         switch (instr.opcode) {
             case Instruction::OpCode::ADD:
             {
+                state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id);
                 for (int i = 0; i < 4; ++i) {
                     if (!swizzle.DestComponentEnabled(i))
                         continue;
@@ -100,6 +108,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
 
             case Instruction::OpCode::MUL:
             {
+                state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id);
                 for (int i = 0; i < 4; ++i) {
                     if (!swizzle.DestComponentEnabled(i))
                         continue;
@@ -113,6 +122,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
             case Instruction::OpCode::DP3:
             case Instruction::OpCode::DP4:
             {
+                state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id);
                 float24 dot = float24::FromFloat32(0.f);
                 int num_components = (instr.opcode == Instruction::OpCode::DP3) ? 3 : 4;
                 for (int i = 0; i < num_components; ++i)
@@ -130,6 +140,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
             // Reciprocal
             case Instruction::OpCode::RCP:
             {
+                state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id);
                 for (int i = 0; i < 4; ++i) {
                     if (!swizzle.DestComponentEnabled(i))
                         continue;
@@ -145,6 +156,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
             // Reciprocal Square Root
             case Instruction::OpCode::RSQ:
             {
+                state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id);
                 for (int i = 0; i < 4; ++i) {
                     if (!swizzle.DestComponentEnabled(i))
                         continue;
@@ -159,6 +171,7 @@ static void ProcessShaderCode(VertexShaderState& state) {
 
             case Instruction::OpCode::MOV:
             {
+                state.debug.max_opdesc_id = std::max<u32>(state.debug.max_opdesc_id, 1+instr.common.operand_desc_id);
                 for (int i = 0; i < 4; ++i) {
                     if (!swizzle.DestComponentEnabled(i))
                         continue;
@@ -212,6 +225,8 @@ OutputVertex RunShader(const InputVertex& input, int num_attributes)
 
     const u32* main = &shader_memory[registers.vs_main_offset];
     state.program_counter = (u32*)main;
+    state.debug.max_offset = 0;
+    state.debug.max_opdesc_id = 0;
 
     // Setup input register table
     const auto& attribute_register_map = registers.vs_input_register_map;
@@ -255,6 +270,9 @@ OutputVertex RunShader(const InputVertex& input, int num_attributes)
     state.call_stack_pointer = &state.call_stack[0];
 
     ProcessShaderCode(state);
+    DebugUtils::DumpShader(shader_memory, state.debug.max_offset, swizzle_data,
+                           state.debug.max_opdesc_id, registers.vs_main_offset,
+                           registers.vs_output_attributes);
 
     DEBUG_LOG(GPU, "Output vertex: pos (%.2f, %.2f, %.2f, %.2f), col(%.2f, %.2f, %.2f, %.2f), tc0(%.2f, %.2f)",
         ret.pos.x.ToFloat32(), ret.pos.y.ToFloat32(), ret.pos.z.ToFloat32(), ret.pos.w.ToFloat32(),